jynx8499

Security engineer by day, developing at night. Zero Trust, vuln assessments, secure SDLC.

[jynx8499 @jynx8499]

strip the cleverness out of an attack and the only variable left is how many times you were willing to ask. 20% to 60% on the same fifteen requests https://t.co/bocZ7tjqYO

[jynx8499 @jynx8499]

#DEFCON for the TraceLabs OSINT Search Party today. Real open missing persons cases, scored on whether what you find is new and usable to the people working them. Aggregators and police blotters get rejected on sight. Points are in whatever nobody bothered to look at.

[jynx8499 @jynx8499]

Idk if it s true or not but i heard that @AnthropicAI developers only writing the loop and let Ai handle the rest.. they even stopped prompting. @bcherny - i get it but i still think human assisted coding necessary. Regardless.

[jynx8499 @jynx8499]

my own stub PDF parser panicked on inputs under 5 bytes because i forgot a bounds check. exactly the bug class i was trying to feel. so i built a fuzzer. https://t.co/XHJcA2Iq6W

[jynx8499 @jynx8499]

the scorer is regex based, fast, ships with default patterns for the common attack styles. give it text, get true/false + which pattern matched. drop-in for red team runs

[jynx8499 @jynx8499]

context: owasp llm01 = prompt injection. attacker hides instructions in user input to override your system prompt. “ignore previous instructions and…” is the textbook example. it’s the #1 risk in the llm top 10 for a reason.

[jynx8499 @jynx8499]

A lot happening just shipped a PromptInjectionScorer to #microsoft Pyrit #redteam #airedteam

[jynx8499 @jynx8499]

Love seeing every single SWE becoming a security engineer! “I used to claude to analyze my code” lol you didnt have to , there are already tools out there having doing your code. I mean good job lol

[Thinkengine Sofia @thinkengineio]

Tried building a security workflow in Notion.

12 tools. 3 dashboards. 1 very tired engineer.

We built Sofia instead.

14ms. One agent. Zero setup.

→ https://t.co/u1Nkg3bc4k

[Thinkengine Sofia @thinkengineio]

80% of employees use unauthorized AI. 86% of security teams can't see it. Shadow AI breach: $670K+ per incident. Sofia Developer doesn't create the shadow. She only analyzes it. No backdoor. No room for one. —> https://t.co/u1Nkg3bc4k

[Thinkengine Sofia @thinkengineio]

Security ops runs on tribal knowledge.

Which server is critical. Which alert has been false for 6 months. Which login at 3am is your team.

MSSPs can't know this. Tools can't either.

Sofia does.

14ms. No dashboards. No tool-hopping.

Try —> https://t.co/u1Nkg3bc4k

[Thinkengine Sofia @thinkengineio]

It understands and takes action.

Most tools: something bad happens → alert → you investigate → you act.

Sofia: threat detected → decision made → action taken.

14ms. No polling. No setup. It just runs.

[jynx8499 @jynx8499]

@XFreeze do you think that other AI companies developing AI to protect the humanity ?? lol

[jynx8499 @jynx8499]

hot take: "AI pentesting" tools aren't using AI to exploit anything. they're wrapping available Kali tools behind an LLM that decides which tool to run next. that's scripting with extra steps, not AI exploitation. source: i've actually tried it. the models refuse. 🤷‍♂️ #shannonai

[jynx8499 @jynx8499]

That’s funny. They call this hacking— running a couple of recon and fuzzing, and it hacked lol funny. I don’t think you would need AI to automate penetration testing, especially that phase…. https://t.co/V32tOSDJ2F

[jynx8499 @jynx8499]

I mean models are trained on human feedback to better understand how this works. Look up what s scheduled reinforcement https://t.co/nIjSubO1kR

[jynx8499 @jynx8499]

Problematic!

[Mario Nawfal @MarioNawfal]

🚨MIT researchers have mathematically proven that ChatGPT’s built-in sycophancy creates a phenomenon they call “delusional spiraling.”

You ask it something, it agrees. You ask again, and it agrees even harder until you end up believing things that are flat-out false and you can’t tell it’s happening.

The model is literally trained on human feedback that rewards agreement.

Real-world fallout includes one man who spent 300 hours convinced he invented a world-changing math formula, and a UCSF psychiatrist who hospitalized 12 patients for chatbot-linked psychosis in a single year.

Source: @heynavtoor

[Elon Musk @elonmusk]

Banger 😂

[dev @zivdotcat]

“babe wake up.”

Claude Code is finally open source

[Chaofan Shou @shoucccc]

Claude code source code has been leaked via a map file in their npm registry!

Code: https://t.co/jBiMoOzt8G